We included 499 participants from the Nautilus Project (ClincalTrials.gov IDs: NCT05307549 and NCT05307575), of whom 373 had post-COVID-19 conditions (PCC) and 126 were healthy controls (HCs). Of the PCC patients, 206 were nonhospitalized (mild PCC), 84 were hospitalized, and 83 were admitted to the ICU. Mild PCC only showed mild COVID-19 symptoms in the acute phase, while hospitalized and ICU PCC had severe complications that needed hospitalization, such as pneumonia. As this was a cross-sectional study, the sample was recruited across 16 hospitals in Spain and Andorra consecutively. It was coordinated by the Consorci Sanitari de Terrassa (Terrassa, Barcelona, Spain). Recruitment was carried out between June 2021 and October 2022.

The inclusion criteria for the PCC group were a confirmed diagnosis of COVID-19 according to the WHO criteria with signs and symptoms of the disease during the acute phase, a period of at least 12 weeks after infection, and age between 18 and 65 years. The exclusion criteria were an established diagnosis of a psychiatric disorder, neurological disorder, neurodevelopmental disorder, or systemic pathology known to cause cognitive deficits before COVID-19 infection and motor or sensory alterations that could interfere with the neuropsychological assessment. The HCs had not had COVID-19 infection (no positive tests or compatible symptoms). The same exclusion criteria for the PCC group were applied to the HC group.

This procedure has been previously described in another study (Carnes-Vendrell et al., 2024). In summary, participation was completely voluntary, and we obtained written informed consent from all the participants before inclusion. We collected data on sociodemographic characteristics, previous comorbidities and COVID-19 symptoms in the first session. At the second visit, the neuropsychological assessment was performed. Different cognitive domains were assessed with an extensive and comprehensive neuropsychological battery. The Spanish version of Rey’s Auditory Verbal Learning Test (RAVLT) (Schmidt, 1996; Alviarez-Schulze et al., 2022) was used to assess verbal memory, whereas the Rey–Osterrieth Complex Figure Test (ROCF) (Meyers and Meyers, 1996) was used for visual memory (immediate and delayed). The copy trial of the ROCF evaluated visuoconstructive abilities. The WAIS-III Digit Span subtest was used to measure verbal attention (digit span forward) and working memory (digit span backward) (Wechsler, 1999). Information processing speed was assessed with the digit symbol (DS) test from the WAIS-III (Wechsler, 1999). Parts A and B of the Trail Making Test (TMT) were administered to measure visual scanning, motor speed and attention, and mental flexibility (Reitan, 1958). Verbal and semantic fluency were assessed with the Controlled Oral Word Association Test (COWAT) (Benton and Hamsher, 1989). However, for verbal fluency, the letters used were P, M and R because Spanish normative data exist for these three tests (Peña-Casanova et al., 2009). The third part of the Stroop (colour-word) test (Golden, 2005) was used as a measure of cognitive inhibitory control (executive functions). To evaluate language, the Boston Naming Test (BNT) was administered (Allegri et al., 1997). Finally, emotion recognition was assessed with the Reading the Mind in the Eye Test (Fernández-Abascal et al., 2013). In addition, information on the cognitive reserve of all participants was collected with the Cognitive Reserve Questionnaire (CRC) (González et al., 2011). All evaluations were performed by trained neuropsychologists.

The participants were given all of the questionnaires to complete online or on paper to assess different variables. In this study, we focused on sleep quality, which was assessed with the Pittsburgh Sleep Quality Index (PSQI) (Buysse et al., 1989); everyday memory failure, which was assessed with the Memory Failures of Everyday Questionnaire (MFE-30) (Sunderland et al., 1984); fatigue, which was assessed with the Chadler Fatigue Questionnaire (CFQ) (Jackson, 2015); and quality of life, which was assessed with the European Quality of Life-5 Dimensions (EQ-5D) (Foundation ER, 2018).

The participants’ anonymity and confidentiality were guaranteed. The Scientific Ethics Committee of the Hospital Universitari Arnau de Vilanova approved both the study and the consent procedure (CEIC 2384), as did the Drug Research Ethics Committee (CEIm) of Consorci Sanitari de Terrassa (CEIm code: 02-20-107-070) and the Ethics Committee of the University of Barcelona (IRB00003099). Additionally, the investigation was conducted in accordance with the latest version of the Declaration of Helsinki.

Descriptive analyses were performed to compare healthy controls with PCC patients. For categorical variables, frequencies and percentages were obtained, and for quantitative variables, the means and standard deviations were obtained. The cognition variables were converted into dichotomous variables depending on whether the result indicated cognitive impairment (−1 standard deviation from the mean), which is why it is shown as a percentage of impairment for each variable. PSQI score of 5 was used as a cutoff indicating good or bad subjective sleep quality (Buysse et al., 1989). Pearson’s nonparametric X² test was conducted, and Fisher’s exact test was used for pairwise differences between groups only for 2×2 tables. Differences in continuous variables between groups were assessed using factorial ANOVA with a general linear model and powerful estimation (robust covariances). All multiple comparisons were adjusted by Bonferroni correction.

To study the relationship between sleep quality and cognitive impairment in PCC and to verify the possible moderating effect of other variables of interest (cognitive reserve, fatigue, subjective cognitive complaints and quality of life), binary logistic regression was applied for each cognitive variable (alteration vs. non-alteration). The explanatory variables were sleep quality (<= 5 good; >5 bad), the possible moderators (MFE-30, CRC, CFQ and EQ_5D) and the 4 interactions between moderators and sleep quality. The models were adjusted for sex, age, and years of schooling and were calculated for PCC participants and healthy controls separately. If the interaction was significant (p < 0.05), then moderation was calculated. Only results in which moderation was present are shown in the results section. The odds ratio (OR) is shown with the 95% CI. For the quality of fit, the Hosmer–Lemeshow test and the area under the curve (AUC) were used.

The statistical significance level used in the analyses was 5% (α = 0.05). All analyses were performed with IBM SPSS statistics 26.

Of the 373 PCC patients, 206 had mild PCC (the ones that had not been hospitalized) (mean age 46.74 years, standard deviation 9.47), 84 were hospitalized (53.07 ± 8.83), and 83 were admitted to the ICU (52.24 ± 8.36). In the mild PCC group, the majority of participants were female (70.1%), while in the hospitalized and ICU groups, the majority of participants were male (51.2 and 53%, respectively). Patients in the ICU-PCC group consumed more alcohol (39.8%), were more likely to be obese (54.2%) and had more previous comorbidities, such as high blood pressure (30.1%) and dyslipidaemia (22.9%), whereas the hospitalized PCC group had more chronic pain (17.3%) and diabetes mellitus (9.5%). Table 1 shows the clinical and sociodemographic characteristics of the sample.

In terms of sleep quality, the PCC participants had worse sleep quality than the healthy control participants (p < 0.001). However, there were no statistically significant differences between the groups based on PCC severity. We found significant differences between the groups (p < 0.001) in the percentage of participants who reported poor quality of sleep (those who obtained ≥5 points on the PSQI). The healthy control group had a lower percentage of responses above 5 on the PSQI (40.2%) than the mild-PCC (77.2%) and hospitalized-PCC (66.7%) groups (Table 1).

Of the moderator variables, we found statistically significant differences between PCC patients and HCs in quality of life, daily memory failure, fatigue and cognitive reserve (Table 2). PCC patients obtained worse results for all of these variables.

Many differences in cognitive performance were found between PCC patients and healthy controls (Table 3). Participants with mild PCC had significantly greater impairment in executive function variables, such as Stroop colour-word (p = 0.006), verbal fluency (p = 0.012) and semantic fluency (p = 0.030), than HCs. However, compared with HCs, hospitalized PCC participants had worse verbal memory [learning (p < 0.001) and delayed recall (p = 0.002)], whereas ICU PCC patients had more deficits in verbal memory with immediate recall (p = 0.001) than the HC. In addition, these participants obtained higher percentages of alterations in attention and information processing speed tasks, such as the digit symbol (p = 0.018), TMT A (p = 0.015) and TMT B (p = 0.002), as well as in working memory (digit span forward, p = 0.018) and emotion recognition (p = 0.001).

To determine whether there were variables that influenced or intervened in the relationship between cognition and sleep quality, we examined whether cognitive reserve, fatigue, daily memory failure and quality of life moderated this relationship. Table 4 and Figure 1 show significant models where only moderation occurred in the relationship between sleep quality and cognition by any of the previously described moderator variables (when the moderator and sleep quality interaction was significant, p < 0.05) (Table 4; Figure 1). Among the possible moderators, fatigue (CFQ) was the only factor that did not affect the cognitive performance of participants with COVID-19 regardless of whether the quality of their sleep was good or poor.

Participants with COVID-19 and poor sleep quality had a 4.3% greater risk of immediate verbal memory deficits (RAVLT immediate recall) and more subjective cognitive complaints (MFE-30) (odds ratio (OR) 1.043; confidence interval (CI) 1.023–1.063). Furthermore, these participants’ risk of immediate verbal memory disorders was multiplied by 2.4 when quality of life (EQ-5D) was low (OR = 0.33; CI = 0.145–0.748).

With delayed visual memory (FCRO delayed recall), people with PCC and poor sleep quality had a 13% lower risk of these alterations with greater cognitive reserve (CRC) (OR = 0.963; CI = 0.929–0.999). If they did not have poor sleep quality, the risk of alterations in delayed visual memory decreased by 9% when cognitive reserve increased.

Finally, with regard to processing speed (DS), people with COVID-19 and poor sleep quality had a 6.7% greater risk of processing speed deficits with more subjective cognitive complaints (MFE) (OR 1.059; CI 1.024–1.096). On the other hand, the risk of slowing processing speed tripled as quality of life decreased (EQ-5D) (OR 0.021; CI 0.003–0.141). If PCC participants did not have poor sleep quality, neither subjective cognitive complaints nor quality of life affected immediate verbal memory performance or processing speed performance.

An analysis of the severity of COVID-19 revealed that participants with mild PCC and poor sleep quality had an 8.5% greater risk of alterations in processing speed with more subjective cognitive complaints (MFE-30) (OR 1.085; CI 1.022–1.152) and a 20% lower likelihood with greater cognitive reserve (CRC) (OR 0.796, CI 0.695–0.913). In both cases, if participants did not have poor sleep quality, the moderator variables did not affect processing speed. Additionally, participants with mild PCC and poor sleep quality tripled (multiplied by 2.9) their risk of alterations in delayed visual memory when they had low quality of life (EQ-5D) (OR 0.267; CI 0.096–0.74). If participants did not have poor quality of sleep, quality of life did not affect visual memory performance. Finally, participants in the ICU-PCC group with poor sleep quality had a 20.4% greater risk of more slowing in processing speed with more subjective cognitive complaints (MFE-30) (OR 1.204; CI 1.036–1.398). Additionally, they had a 65% lower risk of difficulties in the same cognitive domain when they had higher quality of life (EQ-5D) (OR 2.46E-05; CI 4.42E-09-0.137). Again, if the participants in the ICU-PCC group did not have poor sleep quality, subjective cognitive complaints and quality of life did not interfere with processing speed.

Another moderator variable that we analysed was cognitive reserve. We assumed that people with higher levels of cognitive reserve would experience less cognitive impairment due to compensatory mechanisms. We found that PCC participants with poor sleep quality had fewer deficits in visual memory and processing speed when their cognitive reserve was greater. If they did not have poor sleep quality, cognitive performance was not affected by cognitive reserve. To our knowledge, there are no previous findings relating cognitive reserve to sleep quality. However, studies have demonstrated the relationship between cognitive performance and cognitive reserve as a possible predictor of future impairment (Costas-Carrera et al., 2022; Devita et al., 2022; Cavaco et al., 2023).

Subjective cognitive complaints were the last possible moderator variable that interfered with cognitive performance. In this case, PCC participants who reported poor sleep quality had more verbal memory and processing speed deficits when they had more complaints about their cognition. However, if they did not have poor sleep quality, their performance in these cognitive domains was not affected by subjective cognitive complaints. Again, few studies have examined the role of subjective cognitive complaints and sleep quality/cognition. Mantovani et al. (2021) studied the prevalence of myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) in a sample of SARS-CoV-2 survivors and explored features such as clinical, neuropsychiatric, and neuropsychological profiles. They found that the ME/CFS-like group had worse sleep quality, fatigue, pain, depressive symptoms, and subjective cognitive complaints than those without ME/CFS-like symptoms. However, their sample was not comparable to ours because they focused on this syndrome.

A recent meta-analysis revealed that the proportion of participants who experienced fatigue 12 or more weeks after contracting COVID-19 was 0.32 (Ceban et al., 2022), reinforcing previous findings that demonstrated that fatigue is one of the most prevalent symptoms in post-COVID-19 patients. However, our results showed that fatigue was not a moderating variable when PCC participants had poor quality of sleep in terms of cognitive performance in any of the assessed cognitive domains. Numerous studies have shown a relationship between sleep quality and fatigue and between fatigue and cognition, but no previous studies have shown a relationship between these three variables. Margalit et al. (2022) explored the risk factors for long-term COVID-19 and its possible pathophysiology. They concluded that individuals with long-term COVID-19-related fatigue had poorer sleep quality and a greater proportion of subjective cognitive complaints. However, they did not assess cognition with objective measures but rather from a subjective impression of cognitive decline. The strong positive correlation between fatigue and sleep disturbances is not surprising as significant interactions between sleep, fatigue, and the autonomic nervous system have been described (Tanaka et al., 2015). In a multicentre study, the authors identified factors associated with cognitive impairment and fatigue (Hartung et al., 2022). They performed two multivariate analyses, one with potential predictors in the acute phase of COVID-19 and one with potential predictors from the post-COVID-19 period (which included sleep quality). Only the univariate analysis, not the multivariate analysis, revealed a significant difference in sleep quality and cognition. However, it should be noted that these authors assessed cognition only with the MoCA. Fatigue was a significant predictor of sleep problems.

Our findings have potential implications for treatment. We demonstrated that sleep quality may be a trigger for cognitive dysfunction, especially in terms of memory and processing speed, with the moderating effects of quality of life, cognitive reserve, and everyday memory failure. Thus, if sleep quality could be maximized, cognitive impairment in PCC individuals could be reduced. The benefits of different types of interventions for cognitive deficits have been proven. However, we did not find any study that focused on improving sleep quality to minimize its negative effects on other variables.

When interpreting the results, several limitations must be considered. We included only subjective measures of sleep and did not collect information about previous sleep disturbances prior to COVID-19 infection. Instead, we collected information on cognitive performance from an extensive neuropsychological assessment, which enabled us to detect minimal cognitive deficits compared to screening tools. In addition, we have the limitations inherent to a cross-sectional study, such as the impossibility of making causal predictions (cause-effect) and selection bias (as it is a study with consecutive recruitment). The last limitation it is related to the fact that some variables related to the severity of the disease may have not been controlled, like symptoms in the acute phase or its duration. To compensate this, we performed robust statistical analysis to determine the relationships between cognition and sleep quality and between cognition and the moderating effects of other variables, and all the analyses have been adjusted for age, sex, and education.